Selling an AI Software Company

Windsor Drake advises founders of applied AI companies on the sale of their businesses. AI M&A is the most active capability race in technology, and it is also the market where classification does the most work. Two companies with identical revenue can price far apart because one is underwritten as an application business, another as infrastructure, and a third as a talent acquisition. Understanding which transaction you are actually in, and preparing the evidence that supports the best available classification, is the core of sell-side work in this sector. Windsor Drake publishes ongoing AI software valuation research that tracks how these classifications price quarter by quarter.

The layer determines the multiple

Application-layer companies, products that solve a defined workflow with AI inside, are underwritten like vertical software with an AI premium where the AI is demonstrably load-bearing: retention, revenue durability, and workflow ownership set the base, and the model layer must survive the question of what happens when foundation models improve. Infrastructure-layer companies, tooling for data, deployment, evaluation, and serving, are underwritten on developer adoption, consumption economics, and strategic scarcity, with buyers paying for position in the stack rather than end-customer workflow. Talent-led companies, where the asset is the research and engineering team, are underwritten on the team itself, and everything else is diligence.

Founders do not always get to choose their layer, but they choose the evidence. A company presenting itself as an application business without retention cohorts will be priced as talent. A company with genuine infrastructure adoption that markets itself like an app forfeits the scarcity premium. The positioning decision is not spin; it is selecting the underwriting frame the facts genuinely support and then meeting that frame’s evidentiary standard.

Revenue durability is the evidence that moves price

AI revenue attracts a specific skepticism: buyers have seen pilot revenue evaporate, and they underwrite against it. The exhibits that answer the skepticism are concrete. Production deployments rather than pilots, with customers in their second and third renewal. Net revenue retention with expansion driven by usage growth rather than repricing. Gross margins presented honestly against inference and serving costs, with the trajectory as compute costs change. Contracts that survive model substitution, meaning value anchored in workflow, data, and integration rather than access to any particular model. A company that can show two years of cohort retention on production usage has answered the only question that matters; a company that cannot will negotiate against the assumption that revenue is experimental.

The acqui-hire is a real competitive process

When the team is the asset, founders often assume there is nothing to negotiate. The opposite is true. Talent-driven acquisitions are priced against the acquirer’s alternative cost of assembling an equivalent team, and multiple bidders for the same team change that price substantially. Running even a compressed competitive process, with the team’s composition, shipped systems, and research output documented like the asset it is, routinely improves both the aggregate consideration and its split between shareholders and retention packages. The same discipline applies to how security-adjacent AI teams are priced by platform acquirers filling capability gaps.

Frequently asked questions

How much is my AI company worth?

It depends on which layer buyers underwrite you in. Application businesses with durable retention are typically priced as premium vertical software; infrastructure with real adoption is typically priced on strategic scarcity; talent-led companies are priced against the cost of assembling the team. The preparation work is assembling evidence for the strongest classification your facts support.

What do acquirers look for in an applied AI company?

Production deployments rather than pilots, cohort retention through renewals, expansion driven by usage, margins presented net of inference costs, and value that survives model substitution. Buyers also underwrite the team explicitly, and data rights and IP provenance receive close diligence in every AI transaction.

Is an acqui-hire a failure outcome?

No. For talent-led companies it is the correct market, and it is negotiable like any other deal. The aggregate price and its allocation move materially with competitive tension and with how well the team’s output is documented. Treating an acqui-hire as a process rather than a surrender is often worth as much as any product metric.

Will foundation model progress make my company unsellable?

Not if value is anchored where models cannot reach: proprietary workflow position, integration depth, customer data under contract, and distribution. Buyers price model dependence as risk, so the preparation is demonstrating what remains defensible as the model layer commoditizes.